Attached the full output of the test when failing, with Elasticsearch request traces enabled. I suspect the problem is related to purges being executed twice (when we purge a child class first, then the parent class which will trigger a purge on the child class too).